Conversations and Sessions

Sessions give each AI conversation its own transcript while sharing the Task’s files.

What a Session contains

A Session is one continuing conversation within a Task. It has its own transcript, provider and model context, and execution lifecycle. Multiple Sessions can work on the same Task files, which makes implementation and review conversations possible without combining their conversation histories.

Continue or separate

Add a Session to the same Task when you want another conversation over the current shared files. Create a new Task when the work should have a separate file area and branch. A new Session is not a fresh copy of the Task.

History and tabs

Closing a browser tab only hides that view. It does not close the Session or discard Task files. Closing a Session preserves readable history and the Task’s files under the product lifecycle rules. If a request’s delivery status is uncertain after a disconnect, inspect its status before deciding whether to retry; an ambiguous send is not automatically replayed.

Provider-native conversation state stays private to its Session. Other Sessions share the Task files, not that private state.